ABSTRACT
Background: COVID-19 is highly heterogeneous in clinical severity and outcome. Considerable advances have uncovered biomolecular traits associated with fatal outcome. However, novel analytical tools are needed to rapidly and accurately delineate patient subgroups with various immunovirological profiles, analyze diverging disease trajectories and prioritize in-depth molecular studies. Methods: To find how immunovirological features are interrelated, we profiled 12 plasma analytes (SARS-CoV-2 vRNA, SARS-CoV-2-specifc antibodies, cytokine and tissue injury markers) in 500 acute longitudinal plasma samples collected from 214 hospitalized COVID-19 patients. We analyzed them simultaneously using PHATE algorithm (potential of heat diffusion for affinity-based transition embedding, Moon et al, Nature Biotech 2019), which can reduce multiple input variables to two salient features for visualization. We performed whole blood transcriptomic analyses to identify molecular signatures associated with survival vs death in a patient cluster identified as being at extreme mortality risk. Results: PHATE analysis of samples collected 11 days after symptom onset (DSO11) revealed four distinct k-means clusters of patients, which aligned with disease severity and outcome. Two groups were highly enriched in critical patients requiring mechanical ventilation: a high-fatality critical cluster 1 accounted for 59% of fatal outcomes (16/27) by DSO60, while critical cluster 2 had good prognosis. Clusters 3 and 4 consisted almost entirely of non-critical survivors delineated respectively by low and high antibody responses. Averaged trajectories between DSO3 to DSO30 diverged between clusters. All patients of the high-fatality cluster had detectable plasma vRNA, which lingered unlike the critical survivor cluster. Their antibody response had a 4-day delay, while their cytokine profile diverged from the other clusters by DSO8, remaining distinct until DSO22. Transcriptome profiles differed between deceased and survivors of the high-fatality cluster 1, with differential expression of GO terms associated with metabolic processes, protein regulation, cell signaling and immune pathways. Conclusion: This unbiased approach gives an integrated view of dysregulated immune response components in fatal COVID-19, which may be explained through differences in molecular pathways. This approach allows to efficiently target detailed investigations on very high-risk patient subgroups who may most likely benefit from new therapeutic interventions.